An Information Processing Analysis of Product Labels

ABSTRACT - This paper analyzes consumers' use of product label information. Consumers making purchase decisions are assumed to strive to minimize effort as well as to maximize utility. A task analysis of familiar decision strategies exposes both information processing effort and the potential for error in several possible choice environments. The results of the task analysis suggest ways in which product labeling formats can be improved. For example, in certain situations simplified product information that itself is only approximately accurate can lead to more accurate purchase decisions.

Citation:

Michael D. Johnson (1980) ,"An Information Processing Analysis of Product Labels", in NA - Advances in Consumer Research Volume 07, eds. Jerry C. Olson, Ann Abor, MI : Association for Consumer Research, Pages: 724-728.

Advances in Consumer Research Volume 7, 1980     Pages 724-728

AN INFORMATION PROCESSING ANALYSIS OF PRODUCT LABELS

Michael D. Johnson, University of Chicago

ABSTRACT -

This paper analyzes consumers' use of product label information. Consumers making purchase decisions are assumed to strive to minimize effort as well as to maximize utility. A task analysis of familiar decision strategies exposes both information processing effort and the potential for error in several possible choice environments. The results of the task analysis suggest ways in which product labeling formats can be improved. For example, in certain situations simplified product information that itself is only approximately accurate can lead to more accurate purchase decisions.

Groups as diverse as economic theorists and consumer advocates have claimed that providing consumers with more information will result in better point of purchase decisions. Often, however, this belief is based on the assumption that people are able to access and use all the information in the environment in a cost-free way.

Human information processing, however, is not cost-free. If it were, then consumers would use all relevant product label information all the time. Of course, this is not the case. Even the limited empirical evidence available illustrates the neglect of most product label information by consumers (Miller 1978).

The major assertion of this paper is that the benefits obtained from reducing decision-making errors are paid for by increased information processing costs. That is, there is a compensatory trade-off between the effort expended and the accuracy of the choice. The consumer would like to minimize both effort and error, but because these two goals are incompatible consumers must trade off error for effort or effort for error. Such an error/effort framework has been used to describe how people process and use information when making binary choices (Russo and Dosher 1976). In a binary choice task, Russo and Dosher found their subjects adopting simplifying heuristics in order to trade off error for effort.

Because label information is often neglected, it seems reasonable to infer that the disutility of effort is high relative to error. There are two possible explanations. The temporal nature of the process may discount the disutility of error relative to effort. Information processing effort is salient at the point of purchase while feedback as to the quality of the decision is not. Therefore, the disutility of processing costs might be overweighted at the point of purchase.

The salience of processing costs has been used to explain the fact that people often use more than one strategy to eliminate alternatives (Johnson 1979). Elimination type decision heuristics are often replaced by compensatory strategies once all but two or three alternatives have been eliminated from the choice set (Payne 1976). Johnson argues that such behavior is more consistent with the minimization of processing effort than error.

A second explanation places blame on the mode of presentation or format of the product label information. It is possible that more product quality information would be utilized if it were presented in highly processable formats. For example, Russo (1977) has shown that by reorganizing unit prices into a single list consumers buy products with lower unit prices. Thus, the disutility of effort may appear higher because the information is presented in hard-to-use formats.

Indeed, the main purpose of this paper is to analyze alternative label formats from the point of view of the minimization of effort and error. Additionally, this research is meant to answer the question: how can information formats be redesigned to facilitate the use of label information and thereby improve consumer decisions?

THE SCOPE OF THE ANALYSIS

The analysis that follows pertains to naive consumers making product quality decisions at the point of purchase. In this context, a naive consumer is one who has little or no prior information about the product. Differentiation by quality must be based on label information only. Therefore, the scope of the analysis covers retail purchases of consumer durables and nondurables with heterogeneity of quality within a product category.

Error and Effort

Information processing effort is essentially the degree of utilization of attention over time. Acquisition, encoding and comparison of label information all require the use of one's limited information processing capacity (or attention). The approach taken here is to demonstrate how expected effort differs as a function of decision-making strategies and labeling environments.

Where effort is a function of the purchase decision process, error is a function only of the output of that process. Error is the difference in utility between a consumer's best choice alternative and his or her actual choice. The analysis that follows attempts to estimate this error. It identifies and tries to quantify and integrate factors that will inhibit a consumer from making a good purchase decision.

At this point, a simplifying assumption is necessary. Because of the variable nature of the factors affecting both error and effort, further analyses must be based on the expected effort and error. For example, a decision strategy like Tversky's (1972) elimination by aspects rule requires varying amounts of effort to complete the decision, depending on the results of earlier stages in the process. When the EBA rule ignores more information, the likelihood of error increases. Since the absolute amount of information ignored will vary, typical or average values of both effort and error can be obtained only by estimation.

Labeling Environments

Label information will include all attributes of product quality, but will exclude price. Price is a qualitatively distinct attribute dependent on the product/ distributor interaction. The type of label information presented to the consumer will define the labeling environment.

Three major characteristics of the labeling environment are considered. First, labels contain either dichotomous or continuous information. Dichotomous refers to 0-1 or yes-no type of information. For example, a product either has an attribute (or a minimally acceptable level of an attribute) or it does not. Continuous information refers to scales with more than two classifications. Second, labels differ as to whether attribute or summary information is given. Summary information means that relevant attribute information has already been combined or concatenated in a mechanical fashion. For example, many overall evaluations reported in Consumer Reports magazine are obtained by standardizing the attribute scale values and summing them to obtain summary ratings. Third, in a broadly defined sense labels will differ with respect to the organization of the information. The difference is essentially between single product labels and product category lists. A systematic generation of possible labeling environments that differ with respect to these physical properties yields 2³ or eight such environments.

In four of these eight systematically generated environments summary ratings are provided. Summary ratings will be computed using additive models which give each attribute a specific and invariant weight in the overall evaluation. It is quite improbable that the single set of weights and cutoffs assigned to the attributes will reflect the subjective weights and cutoffs of every consumer. Consumers will differentially weight attributes in accord with their subjective importance. As a result, summary ratings can be inaccurate or misleading for a given consumer. In some cases the variability of attribute weights and cutoffs across consumers will be quite large, as for automobiles. In other situations consumers will have more uniform weights which can be accurately captured in a summary rating, as for many small appliances.

The summary ratings are therefore assumed to be either mechanically accurate (MA) or mechanically inaccurate (MI). Although this is actually a continuous variable, the MA/MI dichotomy is used here for simplicity. It is helpful to keep in mind that the mechanical accuracy of summary ratings is unlike the lists versus labels or dichotomous versus continuous variables in that accuracy is not a physical property of the label. It is a more global aspect of the product label environment and the preferences of shoppers.

Other environmental differences have received substantial attention elsewhere and will not be covered here. It will be assumed that all attribute ratings are compatible. That is, all attributes are standardized to use the same rating scale. This removes a source of bias against decision-making strategies which must combine information across attributes. The differences in process caused by the use of numerical, verbal, or pictorial ratings will also be deferred to other discussions. For simplicity, it will be assumed that only numerical ratings are used.

Task Analysis

A task analysis is used to estimate both effort and error. First, it measures expected processing effort expended in a given labeling environment by a consumer using a specific strategy. Second, the task analysis examines factors that increase the expected error when a given decision strategy is being implemented.

When summary ratings are provided, one of two strategies is implied. If the ratings are continuous, the consumer simply chooses the product with the highest rating. If the ratings are dichotomous, the consumer simply chooses the first alternative to receive an acceptable rating (e.g., the Good Housekeeping Seal). When summary ratings are absent, attribute information is provided. Attribute ratings will be assumed to be processed by an consumer using one of four possible choice heuristics. They include an additive utility strategy (AU), an elimination by aspects strategy (EBA), an additive difference strategy (AD), and a conjunctive elimination rule (Conj.). These four choice heuristics were chosen because each represents one of four distinct ways in which attribute information can be searched and integrated (Payne 1976).

The additive utility strategy involves utility calculations for each alternative, while the additive difference strategy involves utility calculations of attribute differences among pairs of alternatives. The EBA and conjunctive rules both involve elimination of alternatives via attribute cutoffs. For a detailed discussion see Bettman (1978).

The nature of the environment and the specific strategy employed determines which of several factors affect expected effort and error. Important factors affecting effort are the number of sources of information searched or referenced, pieces of information searched, pieces of information held in short-term memory (STM), and the required computations or individual operations that must he performed.

Although error is a function of output or the actual choice process, several factors can be isolated as contributors to expected error. They include computational errors, loss of the precision in continuous information due to dichotomization of attribute ratings, mechanical inaccuracy of the summary ratings, loss of information due to the noncompensatory nature of some heuristics, and forgetting due to exceeding STM limitations.

Estimating Error and Effort

Each factor contributing to effort and error is quantified and added to the other factors to obtain estimates of effort and error under all environmental and strategy conditions. The quantification of effort follows directly from normative procedures for using each of the six decision strategies (four choice heuristics and two summary labels). For example, when using a summary label, the number of pieces of information searched is equal to N, the number of alternatives. When using an AU strategy, N ^. M pieces will be searched (where M is the number of attributes).

The quantification of error was calculated after making subjective estimates of the functional relationships between the factors mentioned and expected error. For example, error due to the loss of information because continuous attribute ratings have been dichotomized was assumed to be an increasing linear function of the number of ratings which have been dichotomized (N ^. M) and the degree of dichotomization (r_m - 1, where r_m is the range of attribute scale values in the continuous condition). Therefore, error potential for this factor is equal to N ^. M ^. (r_m - 1).

After quantification, the relevant range of values for each factor was determined. If the range on any given factor contributed disproportionately to overall measures of error or effort it was standardized so as to give that factor a more realistic contribution. That is, the ranges were essentially unit weighted. Finally, the relevant factors affecting a given strategy in a given environment were summed to obtain the expected error and effort. This method was chosen due to the high predictive accuracy of additive models in a wide variety of situations (Einhorn 1977). Recognizing the arbitrary nature of the derived values of effort and error, the relative rankings, not the absolute values of effort and error, will be considered more important.

The reader should be aware that there are several factors that probably affect error and effort but cannot be considered a priori in a task analysis. Such factors include the similarity among the alternatives, the degree of precision with which a heuristic is implemented, and practice effects of using strategies over time. A complete account of the assumptions, procedures, and analysis of individual components of error and effort in the form of a technical appendix can be obtained from the author.

RESULTS

Expected error and effort have been plotted against the number of alternatives in the choice set (where N is set equal to M, the number of attributes, in order to simplify plotting). Figures 1 and 2 show how total effort changes as N and M increase for the continuous list and label conditions. Dichotomization of the information does not significantly alter the results. The results in Figures 1 and 2 indicate that for small N, the summary ratings (MA and MI) and the elimination heuristics are almost effortless. The AU and AD strategies involve slightly more effort. As N and M increase, effort in the AD strategy increases in the most accelerated fashion, followed by the AU and the two elimination strategies respectively. In contrast, the summary rating conditions increase at a small and linear rate so that, as N and M increase, increasing amounts of additional effort are required in order to use the attribute related heuristics. The AU strategy is a little over twice as effortful as summary ratings when N and M equal 4. When M and N equal 8, the difference is more than fourfold.

FIGURE 1

CONTINUOUS LABELS

FIGURE 2

CONTINUOUS LISTS

A further result finds lower effort under every condition in the list environments when compared to the label environments. Interestingly, lists have a greater impact on the EBA strategy than on the conjunctive strategy due to the large reduction in sources of information that need be searched.

Total error was computed for dichotomous and continuous label environments where the attribute information is naturally dichotomous (r_m = 1) and where the attribute information is naturally continuous (r_m = 10). The results are quite interesting. They are presented in Figures 3 through 6.

First of all, when summary ratings are continuous while attribute ratings are not (Figure 3, r_m = 1), the MA and AU strategies involve minimal expected error. And unlike the other conditions, this potential error does not increase significantly as N and M increase. The error rate increases fourfold for MI summary ratings as N and M increase from 4 to 10 while the MA and AU strategies remain essentially constant. As it turns out, this is just the type of situation where the use of MI summary ratings will maximize expected error. This is because when attribute related information is naturally dichotomous, the heuristic strategies are relatively easy and error free.

FIGURE 3

CONTINUOUS INFORMATION

FIGURE 4

DICHOTOMOUS INFORMATION

Second, when summary ratings are forced to be dichotomous (e.g., Good Housekeeping Seal of Approval) and rm = 1, the summary ratings (MI followed by MA) result in greater expected error than any of the attribute related choice strategies as N and M increase (see Figure 4). The additive utility strategy results in minimal potential error that increases only slowly with increased attributes and alternatives. This too is a situation where MI summary information maximizes expected error. When N and M both equal 8, error is seven times greater in the MI condition when compared to the compensatory heuristics (AU and AD).

The story is reversed when attribute ratings are continuous (r_m = 10). The four choice heuristics are now susceptible to implementation errors and to information loss when the environment is dichotomized. When the environment is continuous, the AD and AU strategies quickly become infeasible with respect to expected error as N and M increase (see Figure 5). Summary ratings and elimination strategies yield much lower expected error. Even when N and M both equal 6, such strategies yield only about one-third and one-eighth as much expected error as the AD and AU strategies respectively. When the environment is dichotomous (Figure 6), all four choice heuristics become high in expected error as M and N increase. While error grows rapidly for these strategies, the summary ratings not only begin lower in expected error but they increase at a much smaller rate. When N = M = 10, the summary ratings entail less than one-fourth the expected error as the four decision heuristics.

FIGURE 5

CONTINUOUS INFORMATION

FIGURE 6

DICHOTOMOUS INFORMATION

Total Effort and Error

Some conclusions that can be drawn from the task analysis are summarized below:

1.  In some situations, employing different strategies allows the consumer to trade off error for effort or effort for error. That is, the consumer has a set of nondominated strategies from which to choose. For instance, when summary labels have been dichotomized while attribute information is naturally dichotomous (r_m = 1), MI and MA strategies result in more expected error and less expected effort than the attribute related choice heuristics. In other situations, certain strategies clearly dominate others. When attribute information is naturally continuous (r_m = 10), error and effort are minimized when summary information is used, despite the fact that some of these ratings are mechanically inaccurate.

2.  Lists facilitate the minimization of error and effort under all conditions. This would be classified as a major change in the labeling environment which facilitates the minimization of error and effort.

3.  Dichotomous information should only be presented either if it occurs naturally or when it facilitates concatenation enough to offset any loss of information. In fact, the condition under which total expected error for all 6 strategies combined is minimized is when the attribute ratings are dichotomous (see Figures 3 and 4).

4.  If the disutility of error is much larger than the disutility of effort and a mechanically inaccurate rule is all that is available, then an AU strategy will often be optimal, especially when r_m = 1. This implies that in many situations attribute rather than summary information will best facilitate the minimization of error and effort.

5.  If the disutility of error is relatively small, and only mechanically inaccurate ratings are available, summary ratings will still often be optimal because effort will be minimized. The focus here is on engineering goals. It is possible that even MI ratings are best from an engineering standpoint because overall they should help people make better decisions. In many cases it will be better to get people to use inaccurate ratings than no information at all. In fact, it has yet to be determined as to whether the MI/MA dichotomization is realistic. The power of unit weighted additive models would tend to belittle the significance of this variable. This is just one of the many interesting points which needs to be studied empirically.

Summary and Discussion

The crux of any labeling system should be, quite simply, to provide consumers only with as much information as it takes to make a good decision. The task analysis performed above indicates that summary ratings in list form may be a superior way of making information usable. Consumers are able to make more efficient use of effort, and knowledge limitations can be circumvented.

What is needed, and is the subject of the author's current research, is the development of more complete theories of effort, of error, and of how they interact. In doing so, actual error and dependent variables of effort take the place of the task analysis. At present, both reaction time and pupil dilation are being studied as measures of effort. It is hoped that a weighted function of all the different information processing operations and important environmental factors can accurately and consistently predict the subjective effort required to use different types of label information.

Once these theories have been developed, it will be possible to look at any given point-of-purchase product category and recommend what labeling format will be most appropriate in aiding the consumer to minimize error and effort.

REFERENCES

Bettman, James (1978), An Information Processing Theory of Consumer Choice, Reading, Mass.: Addison-Wesley.

Einhorn, H. (1977), "A Simple Multiattribute Utility Procedure for Evaluation," Behavioral Science, 22, 270-282.

Johnson, Eric (1979), "Deciding How to Decide," paper presented at the Seventh Research Conference on Subjective Probability, Utility and Decision Making, August 27-31, Guteborg, Sweden.

Miller, John (1978), Labeling Research--The State of the Art, Marketing Science Institute.

Payne, John (1976), "Task Complexity and Contingent Processing in Decision Making: An Information Search and Protocol Analysis," Organizational Behavior in Human Performance, 16, 366-387.

Russo, J. E. and Dosher, B. (1976), An Information Processing Analysis of Binary Choice, unpublished manuscript.

Russo, J. E. (1977), "The Value of Unit Price Information,'' Journal of Marketing Research, 14, 193-201.

Tversky, A. (1972), "Elimination by Aspects: A Theory of Choice," Psychological Review, 78, 281-299.

----------------------------------------